Deep Learning-Based Prediction of Tunnel Surrounding Rock Deformation and Collapse Risk Assessment

Prediction of surrounding rock deformation and assessment of collapse risk are crucial prerequisites for ensuring safe tunnel construction.To accurately predict and evaluate deformation and collapse risks of surrounding rock induced by excavation unloading,this study proposes a deep learning-based model for tunnel surrounding rock deformation prediction and collapse risk assessment.The Grey Wolf Optimizer (GWO) algorithm is employed to automatically optimize the hyperparameters of the original Long Short-Term Memory (LSTM) model,thereby improving its prediction accuracy.A shallow-buried tunnel section at DK228+965 of a tunnel along the Lunan High-Speed Railway is selected as a case study.Comparative analysis of the original LSTM and the GWO-LSTM models reveals that the GWO-LSTM model reduces the mean squared error (MSE) by approximately 75.0% and increases the coefficient of determination (R²) by about 12.0%,demonstrating superior accuracy and stability in predicting tunnel surrounding rock deformation.Based on the GWO-LSTM prediction results,real-time assessments of collapse risk at different construction stages are conducted.The analysis indicates that on July 9th,surface areas above the tunnel are prone to hazards such as landslides and roof falls.Therefore,it is recommended to enhance monitoring frequency in these areas.The findings provide valuable guidance for ensuring tunnel construction safety.